nivethasaravanan1206/model-city-point-cloud-registration

Model City Point Cloud Registration Dataset

Dataset overview

The Model City Point Cloud Registration Dataset was created by Nivetha Saravanan as part of the master's thesis Deep Learning-Based Point Cloud Registration in the M.Eng. Autonomous Driving programme at Hochschule Coburg.

The data was collected in the controlled indoor Model City environment at the Hochschule Coburg Kronach campus using a 1:10-scale autonomous model vehicle.

The dataset was developed from 37 ROS 2 bag recordings collected across nine experimental configurations, yielding 64,021 registration pairs after preprocessing.

Sensor setup

The vehicle was equipped with:

• Ouster OS1-070-64U-AX: 64-channel 3D LiDAR
• Intel RealSense D435i: RGB-D camera
• OptiTrack motion-capture system: 6-DoF ground-truth poses
• NVIDIA Jetson: onboard ROS 2 data recording

Repository contents

This repository contains the processed dataset in KITTI-compatible format:

• ouster_kitti_format.zip — processed Ouster LiDAR data
• realsense_kitti_format.zip — processed Intel RealSense RGB-D data

The original raw ROS 2 bag recordings are not included in this repository.

Data preparation

The preparation pipeline included:

1. Point-cloud extraction from ROS 2 bag recordings
2. Sensor and ground-truth timestamp synchronisation
3. Removal of invalid and outlier points
4. Voxel downsampling
5. Source-target registration-pair generation
6. Calculation of relative transformations using OptiTrack poses
7. Conversion into a KITTI-compatible directory structure

Experimental scenarios

The recorded configurations include different motion and environmental conditions, such as:

• Pure translation
• Pure rotation
• Fixed-radius circular motion
• Figure-eight motion
• Combined translation and rotation
• Ramp motion
• Repeated runs
• Occlusion conditions

The dataset supports scenario-level evaluation of registration performance across different motion profiles.

Intended uses

The dataset may be used for research and education involving:

• Pairwise point-cloud registration
• LiDAR and RGB-D registration
• Feature and correspondence learning
• Transfer learning and domain adaptation
• Registration robustness evaluation
• Odometry and SLAM research
• Cross-sensor performance comparison
• Autonomous-driving and robotic-perception research

Limitations

• The dataset was recorded in a controlled indoor 1:10-scale Model City environment.
• Its results should not be interpreted as equivalent to full-scale outdoor autonomous-driving performance.
• The Ouster LiDAR and RealSense RGB-D camera have substantially different fields of view, densities and noise characteristics.
• The published archives contain processed KITTI-format data rather than the original ROS 2 bag recordings.
• The environment does not represent changing weather, real traffic or other uncontrolled outdoor conditions.

Creator and attribution

Dataset creator and curator: Nivetha Saravanan
Programme: M.Eng. Autonomous Driving
Institution: Hochschule Coburg
Year: 2026
Master's thesis: Deep Learning-Based Point Cloud Registration
Thesis supervisor: Prof. Georg Arbeiter

Project code, preprocessing scripts and evaluation pipeline:

https://github.com/nivethasaravanan1206/Deep_Learning-Based_Point_Cloud_Registration

If you use this dataset in research, publications, projects or derived datasets, please credit and cite Nivetha Saravanan.

Citation

@dataset{saravanan2026modelcity,
  author    = {Saravanan, Nivetha},
  title     = {Model City Point Cloud Registration Dataset},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/nivethasaravanan1206/model-city-point-cloud-registration}
}

Licence

This dataset is released under the Creative Commons Attribution 4.0 International Licence (CC BY 4.0).

Users may share and adapt the dataset, including for commercial purposes, provided that appropriate credit is given to Nivetha Saravanan, a link to the original dataset is included, and any modifications are identified.

Licence details: https://creativecommons.org/licenses/by/4.0/